One-Way vs Two-Way ANOVA: When Do You Need a Second Factor?

A nursing researcher collects exam scores from 120 students taught by three different methods: traditional lecture, simulation-based learning, and blended. She runs a one-way ANOVA and finds a significant difference, F(2, 117) = 4.83, p = .010. Done? Not quite. Her supervisor points out that half the students are undergraduates and half are postgraduates, and the teaching methods might affect these two groups differently. That question requires a second factor and a different analysis. The one-way ANOVA cannot answer it, because it has no way to detect whether the effect of teaching method changes depending on student level. That is exactly what a two-way ANOVA was built for.

One-Way ANOVA: One Factor, Three or More Groups

One-way ANOVA tests whether the means of a continuous outcome differ across the levels of a single categorical factor. The word "one-way" refers to one independent variable. If a dietitian measures weight loss in 90 participants randomly assigned to three diet plans (keto, Mediterranean, low-fat), that is a one-way design with one factor (diet type) and three levels.

The test produces a single F-statistic. In SPSS, you will find it in the ANOVA table with one row for "Between Groups" and one row for "Within Groups." A result like F(2, 87) = 5.14, p = .008 tells you that at least one diet group lost a different amount of weight than the others. It does not say which group. That is what post-hoc tests like Tukey's HSD are for.

One-way ANOVA works well when your research question involves a single grouping variable. But many study designs include two. A clinical trial might test three drug dosages across male and female patients. A marketing study might compare four ad formats across two age brackets. Ignoring the second factor does not make it disappear. It just hides in the error term, reducing your statistical power and potentially masking real patterns in the data.

Two-Way ANOVA: Two Factors and Their Interaction

Two-way ANOVA (also called factorial ANOVA) tests two independent variables simultaneously. Return to the nursing example: Factor A is teaching method (3 levels), Factor B is student level (2 levels: undergraduate, postgraduate). That gives you a 3 × 2 factorial design with six cells. Each cell represents a specific combination, such as "undergraduates taught by simulation."

The analysis tests three things at once. First, the main effect of Factor A: do exam scores differ across teaching methods, averaged over student levels? Second, the main effect of Factor B: do undergraduates and postgraduates score differently, averaged over teaching methods? Third, and this is where two-way ANOVA earns its place, the interaction effect (A × B): does the effect of teaching method depend on whether the student is an undergraduate or a postgraduate?

With 120 students in a 3 × 2 design and 20 per cell, the SPSS output might show: teaching method F(2, 114) = 5.27, p = .007, η_p² = .085; student level F(1, 114) = 3.91, p = .050, η_p² = .033; and interaction F(2, 114) = 4.12, p = .019, η_p² = .067. All three effects are tested in the same model, and each has its own row in the output table.

What an Interaction Effect Actually Means

The interaction is the reason you run a two-way ANOVA instead of two separate one-way ANOVAs. A significant interaction tells you that the effect of one factor is not the same at every level of the other factor. The lines on an interaction plot cross or diverge instead of running parallel.

Here is a concrete example. Suppose simulation-based learning raises exam scores by 12 points for postgraduates but only 2 points for undergraduates, compared to traditional lecture. Meanwhile, blended learning helps both groups equally, raising scores by about 6 points regardless of student level. The benefit of simulation depends on who is receiving it. That dependency is the interaction.

You can spot this on an interaction plot (called a profile plot in SPSS). Plot teaching method on the x-axis and mean exam score on the y-axis, with separate lines for undergraduates and postgraduates. If the lines are roughly parallel, there is no interaction. If they cross or fan out, an interaction is present. In the example above, the postgraduate line would climb steeply from lecture to simulation, while the undergraduate line would stay nearly flat. That non-parallel pattern matches the significant F(2, 114) = 4.12 for the interaction term.

Main Effects vs. Interaction: Why Order Matters

When the interaction is significant, the main effects can be misleading. Consider why. The main effect of teaching method averages across both student levels. If simulation helps postgraduates enormously but barely moves the needle for undergraduates, the average improvement looks moderate. You might report "simulation had a moderate effect" and miss the real story: it had a large effect for one group and almost none for the other.

The rule is straightforward. Check the interaction first. If it is not significant, interpret the main effects directly and run post-hoc tests (Tukey, Bonferroni, or Games-Howell depending on your variance situation) on whichever main effect is significant. If the interaction is significant, do not interpret the main effects in isolation. Instead, run a simple effects analysis. This means testing the effect of teaching method separately for undergraduates and separately for postgraduates. In SPSS, you do this through the syntax editor using the /EMMEANS subcommand with COMPARE, or by splitting the file by student level and running one-way ANOVAs within each group.

We see this mistake regularly in thesis drafts. A student reports both main effects and the interaction, all in one paragraph, without noting that the significant interaction makes the main-effect averages misleading. Reviewers flag this because it suggests the student does not understand what the interaction means. For guidance on reporting effect sizes alongside these results, partial eta-squared (η_p²) is the standard measure for each effect in the ANOVA table.

Reading the SPSS Output

In SPSS, two-way ANOVA is run through Analyze > General Linear Model > Univariate. Move your dependent variable into the "Dependent Variable" box and both factors into the "Fixed Factor(s)" box. SPSS automatically generates the interaction term.

The output table is called "Tests of Between-Subjects Effects." It has several rows. "Corrected Model" shows the overall fit of the model. Below that, you will find one row for Factor A (e.g., Teaching_Method), one row for Factor B (e.g., Student_Level), and one row for the interaction (Teaching_Method * Student_Level). Each row shows the Type III Sum of Squares, df, Mean Square, F, Sig. (the p-value), and Partial Eta Squared. The "Error" row at the bottom contains the residual variance, and "Corrected Total" is the total variance in the dependent variable.

Partial eta-squared appears in the last column. For the interaction in our nursing example, η_p² = .067 means the interaction accounts for about 6.7% of the variance in exam scores after removing the variance explained by the other effects. Using the standard benchmarks (.01 small, .06 medium, .14 large), that is a medium effect. Report it in APA format as: F(2, 114) = 4.12, p = .019, η_p² = .067.

One-Way vs. Two-Way: Quick Comparison

Feature	One-Way ANOVA	Two-Way ANOVA
Independent variables	1 factor	2 factors
Hypotheses tested	1 (difference among group means)	3 (main effect A, main effect B, A × B interaction)
SPSS output rows	Between Groups, Within Groups	Factor A, Factor B, A*B, Error
Interaction effect	Cannot be tested	Tested directly
Post-hoc tests	On the single factor	On main effects (if no interaction) or simple effects (if interaction is significant)
When to use	One grouping variable, 3+ levels	Two grouping variables, and you suspect their combined effect matters
Example design	3 diet plans → weight loss	3 diet plans × 2 genders → weight loss

The Common Mistake: Two Separate One-Way ANOVAs

This is the single most frequent error we encounter in factorial designs. A student has two factors but runs two separate one-way ANOVAs instead of one two-way ANOVA. First they test teaching method on exam scores, ignoring student level. Then they test student level on exam scores, ignoring teaching method. They report both and call it a day.

This approach has two problems. First, it cannot detect the interaction. If the effect of teaching method differs between undergraduates and postgraduates, two separate one-way ANOVAs will never reveal that. Second, it inflates the error term. When you ignore a factor that actually explains some variance, that variance gets absorbed into the residual. The denominator of the F-ratio grows larger, the F-value shrinks, and real effects become harder to detect. A two-way ANOVA with both factors in the model is more powerful than two one-way ANOVAs run separately, because it partitions the variance more accurately.

If you have two categorical independent variables and one continuous dependent variable, the correct test is a two-way ANOVA. Period. Running it is straightforward in SPSS, and Academic Stats Agent will select it automatically when you describe your variables and upload your data.